search

probml / pyprobml

Python code for "Probabilistic Machine learning" book by Kevin Murphy
MIT License
6.54k stars 1.54k forks source link

make pedagogical JAX EBM demos #674

Closed murphyk closed 2 years ago

murphyk commented 2 years ago

The chapter on energy based models currently only has one demo , which does score matching on 2d swiss roll data. We need more pedagocial examples, eg of contrastive divergence learning, or fitting EBMs to MNIST. Maybe translate the Learnergy pytorch code to JAX?

murphyk commented 2 years ago

See ch 25 of vol 2 for some details.

arpitvaghela commented 2 years ago

@murphyk I would like to take this issue up.

gileshd commented 2 years ago

If it wouldn't be stepping on anyone's toes, I'd like to give this a go too. I was thinking of starting with a simple example of contrastive divergence for a RBM, time permitting, this might be extended to using persistent CD for pre-training a DBM.

maheswarantp commented 2 years ago

@gileshd I am also planning to work on this. But will work on something other than Contrastive Divergence for RBM. Just keeping you in the loop. At the moment, working on fitting EBMs to MNIST.